Multivariate cluster dispersion-covariance matrix effects package

Question

I would like to build the effect of interacting with the error strip using the effects package. However, I want to calculate errors based on a multi-way cluster covariance matrix, which can be calculated using the multiwayvcov package. Effects can take a function to compute the covariance matrix (vcov. =), But it looks like the function will not accept any additional arguments.

Example:

library(effects)
library(ggplot2)
library(multiwayvcov)
m1 <- lm(mpg ~ hp + wt + hp:wt, data=mtcars)
tmp <- as.data.frame(effect("hp:wt", model, vcov=vcov, se=TRUE, xlevels=list(wt=c(2.2,3.2,4.2))))
ggplot(data=tmp, aes(x=hp, y=fit, colour=as.factor(wt))) +
   geom_line() +
   geom_ribbon(aes(ymin=lower, ymax=upper, fill=as.factor(wt)), alpha = 0.5) +
   labs(colour="wt")

The above gives a similar graph for interactions using non-clustered se. I want something that does the same thing, but with a clustered one. Something like (imagine that “gear” and “cylinder” are cluster identifiers in mtcars data):

m1 <- lm(mpg ~ hp + wt + hp:wt, data=mtcars)
cluster.ids <- data.frame(i = mtcars$gear, j = mtcars$cyl)
tmp <- as.data.frame(effect("hp:wt", m1, vcov=cluster.vcov(m1, cluster = cluster.ids), xlevels=list(wt=c(2.2,3.2,4.2))))

Any help is much appreciated

Answer 1

, . , vcov lm ( ) . , , .

:

m1 <- lm(mpg ~ hp + wt + hp:wt, data=mtcars)
tmp1 <- as.data.frame(effect("hp:wt", m1, vcov=vcov, se=TRUE,  xlevels=list(wt=c(2.2,3.2,4.2))))

ggplot(data=tmp, aes(x=hp, y=fit, colour=as.factor(wt))) +
geom_line() +
geom_ribbon(aes(ymin=lower, ymax=upper, fill=as.factor(wt)), alpha = 0.5) +
labs(colour="wt")

hccm covariance matrix ( ) :

library(car)
m2 <- lm(mpg ~ hp + wt + hp:wt, data=mtcars)
hccmfunc <- function(x) {  
return(hccm(x, type="hc0"))}
tmp2 <- as.data.frame(effect("hp:wt", m2, vcov=hccmfunc,   xlevels=list(wt=c(2.2,3.2,4.2))))

, , ( , mtcars):

m3 <- lm(mpg ~ hp + wt + hp:wt, data=mtcars)
cluster.ids <- data.frame(i = mtcars$gear, j = mtcars$cyl)
mwfunc <- function(x) {return(cluster.vcov(x, cluster= cluster.ids))}
tmp <- as.data.frame(effect("hp:wt", m3, vcov=mwfunc, xlevels=list(wt=c(2.2,3.2,4.2))))

NaNs, .